llvm-project/clang/test/OpenMP/taskloop_reduction_codegen.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

225 lines
13 KiB
C++
Raw Normal View History

// RUN: %clang_cc1 -fopenmp -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck --check-prefix SIMD-ONLY0 %s
// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
// expected-no-diagnostics
// CHECK: [[RED_SIZE1:@reduction_size[.].+]] = common thread_local global i64 0
// CHECK: [[RED1:@reduction[.].+]] = common thread_local global i8* null
// CHECK: [[RED_SIZE2:@reduction_size[.].+]] = common thread_local global i64 0
struct S {
float a;
S() : a(0.0f) {}
~S() {}
};
#pragma omp declare reduction(+:S:omp_out.a += omp_in.a) initializer(omp_priv = omp_orig)
float g;
int a;
#pragma omp threadprivate(a)
int main (int argc, char *argv[])
{
int i, n;
float a[100], b[100], sum, e[argc + 100];
S c[100];
float &d = g;
/* Some initializations */
n = 100;
for (i=0; i < n; i++)
a[i] = b[i] = i * 1.0;
sum = 0.0;
#pragma omp taskloop reduction(+:sum, c[:n], d, e)
for (i=0; i < n; i++) {
sum = sum + (a[i] * b[i]);
c[i].a = i*i;
d += i*i;
e[i] = i;
}
}
// CHECK-LABEL: @main(
// CHECK: [[RETVAL:%.*]] = alloca i32,
// CHECK: [[ARGC_ADDR:%.*]] = alloca i32,
// CHECK: [[ARGV_ADDR:%.*]] = alloca i8**,
// CHECK: [[I:%.*]] = alloca i32,
// CHECK: [[N:%.*]] = alloca i32,
// CHECK: [[A:%.*]] = alloca [100 x float],
// CHECK: [[B:%.*]] = alloca [100 x float],
// CHECK: [[SUM:%.*]] = alloca float,
// CHECK: [[SAVED_STACK:%.*]] = alloca i8*,
// CHECK: [[C:%.*]] = alloca [100 x %struct.S],
// CHECK: [[D:%.*]] = alloca float*,
// CHECK: [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]],
// CHECK: [[DOTRD_INPUT_:%.*]] = alloca [4 x %struct.kmp_task_red_input_t],
// CHECK: alloca i32,
// CHECK: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32,
// CHECK: [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32,
// CHECK: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
// CHECK: store i32 0, i32* [[RETVAL]],
// CHECK: store i32 [[ARGC:%.*]], i32* [[ARGC_ADDR]],
// CHECK: store i8** [[ARGV:%.*]], i8*** [[ARGV_ADDR]],
// CHECK: [[TMP1:%.*]] = load i32, i32* [[ARGC_ADDR]],
// CHECK: [[ADD:%.*]] = add nsw i32 [[TMP1]], 100
// CHECK: [[TMP2:%.*]] = zext i32 [[ADD]] to i64
// CHECK: [[VLA:%.+]] = alloca float, i64 %
// CHECK: call void @__kmpc_taskgroup(%struct.ident_t*
// CHECK-DAG: [[TMP21:%.*]] = bitcast float* [[SUM]] to i8*
// CHECK-DAG: store i8* [[TMP21]], i8** [[TMP20:%[^,]+]],
// CHECK-DAG: [[TMP20]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T:%.+]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_:%.+]], i32 0, i32 0
// CHECK-DAG: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 1
// CHECK-DAG: store i64 4, i64* [[TMP22]],
// CHECK-DAG: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 2
// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT1:.+]] to i8*), i8** [[TMP23]],
// CHECK-DAG: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 3
// CHECK-DAG: store i8* null, i8** [[TMP24]],
// CHECK-DAG: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 4
// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB1:.+]] to i8*), i8** [[TMP25]],
// CHECK-DAG: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 5
// CHECK-DAG: [[TMP27:%.*]] = bitcast i32* [[TMP26]] to i8*
// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP27]], i8 0, i64 4, i1 false)
// CHECK-DAG: [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 0
// CHECK-DAG: [[LB_ADD_LEN:%.*]] = add nsw i64 -1, %
// CHECK-DAG: [[ARRAYIDX6:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 [[LB_ADD_LEN]]
// CHECK-DAG: [[TMP31:%.*]] = bitcast %struct.S* [[ARRAYIDX5]] to i8*
// CHECK-DAG: store i8* [[TMP31]], i8** [[TMP28:%[^,]+]],
// CHECK-DAG: [[TMP28]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4:%.+]], i32 0, i32 0
// CHECK-DAG: [[TMP32:%.*]] = ptrtoint %struct.S* [[ARRAYIDX6]] to i64
// CHECK-DAG: [[TMP33:%.*]] = ptrtoint %struct.S* [[ARRAYIDX5]] to i64
// CHECK-DAG: [[TMP34:%.*]] = sub i64 [[TMP32]], [[TMP33]]
// CHECK-DAG: [[TMP35:%.*]] = sdiv exact i64 [[TMP34]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
// CHECK-DAG: [[TMP36:%.*]] = add nuw i64 [[TMP35]], 1
// CHECK-DAG: [[TMP37:%.*]] = mul nuw i64 [[TMP36]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
// CHECK-DAG: store i64 [[TMP37]], i64* [[TMP38:%[^,]+]],
// CHECK-DAG: [[TMP38]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 1
// CHECK-DAG: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 2
// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT2:.+]] to i8*), i8** [[TMP39]],
// CHECK-DAG: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 3
// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_FINI2:.+]] to i8*), i8** [[TMP40]],
// CHECK-DAG: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 4
// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB2:.+]] to i8*), i8** [[TMP41]],
// CHECK-DAG: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 5
// CHECK-DAG: store i32 1, i32* [[TMP42]],
// CHECK-DAG: [[TMP44:%.*]] = load float*, float** [[D]],
// CHECK-DAG: [[TMP45:%.*]] = bitcast float* [[TMP44]] to i8*
// CHECK-DAG: store i8* [[TMP45]], i8** [[TMP43:%[^,]+]],
// CHECK-DAG: [[TMP43]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7:%.+]], i32 0, i32 0
// CHECK-DAG: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 1
// CHECK-DAG: store i64 4, i64* [[TMP46]],
// CHECK-DAG: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 2
// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT3:.+]] to i8*), i8** [[TMP47]],
// CHECK-DAG: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 3
// CHECK-DAG: store i8* null, i8** [[TMP48]],
// CHECK-DAG: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 4
// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB3:.+]] to i8*), i8** [[TMP49]],
// CHECK-DAG: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 5
// CHECK-DAG: [[TMP51:%.*]] = bitcast i32* [[TMP50]] to i8*
// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP51]], i8 0, i64 4, i1 false)
// CHECK-DAG: [[TMP53:%.*]] = bitcast float* [[VLA]] to i8*
// CHECK-DAG: store i8* [[TMP53]], i8** [[TMP52:%[^,]+]],
// CHECK-DAG: [[TMP52]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8:%.+]], i32 0, i32 0
// CHECK-DAG: [[TMP54:%.*]] = mul nuw i64 [[TMP2]], 4
// CHECK-DAG: [[TMP55:%.*]] = udiv exact i64 [[TMP54]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
// CHECK-DAG: store i64 [[TMP54]], i64* [[TMP56:%[^,]+]],
// CHECK-DAG: [[TMP56]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 1
// CHECK-DAG: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 2
// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT4:.+]] to i8*), i8** [[TMP57]],
// CHECK-DAG: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 3
// CHECK-DAG: store i8* null, i8** [[TMP58]],
// CHECK-DAG: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 4
// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB4:.+]] to i8*), i8** [[TMP59]],
// CHECK-DAG: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 5
// CHECK-DAG: store i32 1, i32* [[TMP60]],
// CHECK-DAG: [[DOTRD_INPUT_GEP_]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64
// CHECK-DAG: [[DOTRD_INPUT_GEP_4]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64
// CHECK-DAG: [[DOTRD_INPUT_GEP_7]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64
// CHECK-DAG: [[DOTRD_INPUT_GEP_8]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64
// CHECK: [[TMP61:%.*]] = bitcast [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]] to i8*
// CHECK: [[TMP62:%.*]] = call i8* @__kmpc_task_reduction_init(i32 [[TMP0]], i32 4, i8* [[TMP61]])
// CHECK: [[TMP63:%.*]] = load i32, i32* [[N]],
// CHECK: store i32 [[TMP63]], i32* [[DOTCAPTURE_EXPR_]],
// CHECK: [[TMP64:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]],
// CHECK: [[SUB:%.*]] = sub nsw i32 [[TMP64]], 0
// CHECK: [[SUB10:%.*]] = sub nsw i32 [[SUB]], 1
// CHECK: [[ADD11:%.*]] = add nsw i32 [[SUB10]], 1
// CHECK: [[DIV:%.*]] = sdiv i32 [[ADD11]], 1
// CHECK: [[SUB12:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK: store i32 [[SUB12]], i32* [[DOTCAPTURE_EXPR_9]],
// CHECK: [[TMP65:%.*]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i32 1, i64 888, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @[[TASK:.+]] to i32 (i32, i8*)*))
Do not always request an implicit taskgroup region inside the kmpc_taskloop function Summary: For the following code: ``` int i; #pragma omp taskloop for (i = 0; i < 100; ++i) {} #pragma omp taskloop nogroup for (i = 0; i < 100; ++i) {} ``` Clang emits the following LLVM IR: ``` ... call void @__kmpc_taskgroup(%struct.ident_t* @0, i32 %0) %2 = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 %0, i32 1, i64 80, i64 8, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @.omp_task_entry. to i32 (i32, i8*)*)) ... call void @__kmpc_taskloop(%struct.ident_t* @0, i32 %0, i8* %2, i32 1, i64* %8, i64* %9, i64 %13, i32 0, i32 0, i64 0, i8* null) call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 %0) ... %15 = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 %0, i32 1, i64 80, i64 8, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates.1*)* @.omp_task_entry..2 to i32 (i32, i8*)*)) ... call void @__kmpc_taskloop(%struct.ident_t* @0, i32 %0, i8* %15, i32 1, i64* %21, i64* %22, i64 %26, i32 0, i32 0, i64 0, i8* null) ``` The first set of instructions corresponds to the first taskloop construct. It is important to note that the implicit taskgroup region associated with the taskloop construct has been materialized in our IR: the `__kmpc_taskloop` occurs inside a taskgroup region. Note also that this taskgroup region does not exist in our second taskloop because we are using the `nogroup` clause. The issue here is the 4th argument of the kmpc_taskloop call, starting from the end, is always a zero. Checking the LLVM OpenMP RT implementation, we see that this argument corresponds to the nogroup parameter: ``` void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t *task, int if_val, kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int nogroup, int sched, kmp_uint64 grainsize, void *task_dup); ``` So basically we always tell to the RT to do another taskgroup region. For the first taskloop, this means that we create two taskgroup regions. For the second example, it means that despite the fact we had a nogroup clause we are going to have a taskgroup region, so we unnecessary wait until all descendant tasks have been executed. Reviewers: ABataev Reviewed By: ABataev Subscribers: rogfer01, cfe-commits Differential Revision: https://reviews.llvm.org/D53636 llvm-svn: 345180
2018-10-25 03:06:37 +08:00
// CHECK: call void @__kmpc_taskloop(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i8* [[TMP65]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 1, i32 0, i64 0, i8* null)
// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t*
// CHECK: ret i32
// CHECK: define internal void @[[RED_INIT1]](i8* %0)
// CHECK: store float 0.000000e+00, float* %
// CHECK: ret void
// CHECK: define internal void @[[RED_COMB1]](i8* %0, i8* %1)
// CHECK: fadd float %
// CHECK: store float %{{.+}}, float* %
// CHECK: ret void
// CHECK: define internal void @[[RED_INIT2]](i8* %0)
// CHECK: load i8*, i8** [[RED1]],
// CHECK: call void [[OMP_INIT1:@.+]](
// CHECK: ret void
// CHECK: define internal void [[OMP_COMB1:@.+]](%struct.S* noalias %0, %struct.S* noalias %1)
// CHECK: fadd float %
// CHECK: define internal void [[OMP_INIT1]](%struct.S* noalias %0, %struct.S* noalias %1)
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(
// CHECK: define internal void @[[RED_FINI2]](i8* %0)
// CHECK: load i64, i64* [[RED_SIZE1]]
// CHECK: call void @
// CHECK: ret void
// CHECK: define internal void @[[RED_COMB2]](i8* %0, i8* %1)
// CHECK: load i64, i64* [[RED_SIZE1]]
// CHECK: call void [[OMP_COMB1]](
// CHECK: ret void
// CHECK: define internal void @[[RED_INIT3]](i8* %0)
// CHECK: store float 0.000000e+00, float* %
// CHECK: ret void
// CHECK: define internal void @[[RED_COMB3]](i8* %0, i8* %1)
// CHECK: fadd float %
// CHECK: store float %{{.+}}, float* %
// CHECK: ret void
// CHECK: define internal void @[[RED_INIT4]](i8* %0)
// CHECK: load i64, i64* [[RED_SIZE2]]
// CHECK: store float 0.000000e+00, float* %
// CHECK: ret void
// CHECK: define internal void @[[RED_COMB4]](i8* %0, i8* %1)
// CHECK: load i64, i64* [[RED_SIZE2]]
// CHECK: fadd float %
// CHECK: store float %{{.+}}, float* %
// CHECK: ret void
// CHECK: call i8* @__kmpc_task_reduction_get_th_data(
// CHECK: call i8* @__kmpc_task_reduction_get_th_data(
// CHECK: call i8* @__kmpc_task_reduction_get_th_data(
// CHECK: call i8* @__kmpc_task_reduction_get_th_data(
// CHECK-DAG: distinct !DISubprogram(linkageName: "[[TASK]]", scope: !
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT1]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB1]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT2]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_FINI2]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB2]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT3]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB3]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT4]]"
// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB4]]"